Notes from the
grid frontier.
Engineering, product, and industry deep-dives from the SmartTec team.
The research is in: AI data centers need batteries at every layer
A 2026 wave of peer-reviewed work — from a Nature Energy paper to arXiv reviews — converges on one finding: AI power profiles break traditional data center design, and layered energy storage is the fix.
All posts
AMD’s Helios moment: what Microsoft’s rack-scale bet means for a 30-GPU operator
Microsoft just became the first hyperscaler to commit to AMD’s Helios racks at scale — 72 MI455X GPUs, 31.1 TB of HBM4, inference-first. Here’s the honest read from an operator whose Phase 1 is contractually NVIDIA.
Where a 110 kW operator fits in a 3.5 GW world
CoreWeave and Nebius have contracted ~3.5 gigawatts each. The Gulf is deploying sovereign AI by the hundreds of megawatts. So what is a 30-GPU site in Oklahoma doing in this market? Winning different rows of the table.
What China’s underwater data centers teach a landlocked Oklahoma operator
The world’s first commercial subsea data center has run 35 meters under the South China Sea for nearly three years. We operate on dry land in Mead, Oklahoma — and we’re taking notes anyway.
Korea builds gigawatts, Japan builds pods — and what 110 kW in Oklahoma has in common with Osaka
Asia’s two AI-infrastructure superpowers picked opposite strategies. Korea is pouring ~550 trillion won into 18.4 GW of national AI campuses; Japan is scattering liquid-cooled GPU pods through its cities. Both are right — for their constraints.
The thermal wall: what peer-reviewed cooling research says about your next GPU cluster
Air cooling is hitting a physical ceiling just as the average rack jumped 69% in a year to 27 kW and single GPUs crossed 1,000 watts. Here’s what the 2025–26 research literature says each cooling method can actually hold.
PUE is money: the real economics of cooling choices
PUE 1.45 versus 1.06 sounds like a rounding error until you multiply by 8,760 hours. On one megawatt of IT load, the gap between air and immersion cooling is a quarter-million dollars a year — before you count the throughput.
Why we're building a battery-backed AI cloud
AI workloads need 10x the power of traditional compute. The grid can't keep up — interconnection queues are 4 to 7 years long. We bet that owning the power layer is the only way to ship AI compute at the pace the market needs.
Announcing our design partner program
Three slots, locked pricing for 12 months, direct engineering access, named case study at power-on. Here's how we're picking our first three production customers.
Cerebras vs. NVIDIA H200: when to use which
A practical guide for inference teams choosing between NVIDIA H200 and Cerebras CS-3. The answer is rarely "one or the other" — most workloads benefit from running both.
How AURA rides through grid events
AURA watches ISO/RTO load forecasts, weather, and historical event patterns to predict grid instability before it happens. Here's how the model works and what it caught in our first 90 days.
The AI infrastructure build-out is bottlenecked on power, not chips
Everyone is talking about GPU shortages. The real bottleneck is megawatt-scale power delivery. We pulled 18 months of data on interconnection queues and the picture is grim.
What a 5 MW AI compute hall actually costs
Capex, opex, PPA, lease — we break down the real numbers for a 5 MW AI compute hall running NVIDIA H100s on z1power battery-backed power.
Sub-10ms failover: how we keep GPUs running through grid events
When the grid drops, our batteries take over in under 10 milliseconds. Here's the switchgear, the controls, and the load-shedding logic that makes it work without interrupting your training job.
Behind the scenes: building the SmartTec brand
Why we refreshed the brand, what we kept, what we threw away, and how the design system ended up looking like a battery manufacturer's spec sheet (in the best way).
The economics of behind-the-meter AI compute
When you own the power, your cost structure changes. We model out five years of TCO for grid-tied vs. behind-the-meter AI compute at 1 MW, 5 MW, and 50 MW scales.
Why we're starting with NVIDIA Cloud Partner architecture
Reference architectures aren't glamorous, but they save months of integration work. Here's why we picked NVIDIA's NCP framework for our base deployment and what it unlocks for customers.
Announcing SmartTec: the battery-backed AI cloud
Today we're announcing SmartTec. NVIDIA and Cerebras compute on megawatt batteries we build ourselves. Q4 2026 power-on. Three design-partner slots now open.